UC Berkeley

My work contributes to three related goals in neuroscience: to understand the relationship between plasticity and learning, to elucidate the dynamics between the brain and behavior during learning, and to infer how the coordinated activities of ensembles of neurons mediate behavior. My thesis is divided into three main chapters reflecting these goals. Each chapter addresses a different aspect of the question of how changes in the brain manifest as changes in behavior. I have attempted to build a bridge between physiological, behavioral, and computational approaches to the neuroscientific study of learning. The anchor that motivates my entire thesis is an inquiry into learning behavior.

First, I present a system developed collaboratively in our laboratory for providing real-time feedback to the brain in the form of microstimulation. We demonstrate the utility of this system for introducing microstimulation into the brain conditional upon the action potentials of neurons, the phase of local field potentials, or behavioral events. This engineering project presents a system made out of off-the-shelf components for using cortical microstimulation to interface with the brain on a time-scale sufficient for engaging spike timing-dependent plasticity in behaving animals.

Second, I detail a rodent animal model of learning behavior involving the use of intra-cortical microstimulation (ICMS) within the primary somatosensory barrel fields (S1bf). This paradigm is novel both in its use of ICMS to bypass feedforward input from the sensory periphery, and in the approach I adopted toward the subjects' learning behavior. Instead of simply measuring learning behavior as some average learning rate across subjects, the focus is upon the learning behavior of the individual subjects. This perspective revealed a tight correlation between distinct behaviors observed during the learning process and changes in the neural response of S1bf to ICMS. These changes manifested as an increase in the duration of the inhibitory response to ICMS as each subject began to respond, as well as an increase in the excitatory response to ICMS as the subjects' consolidated their learning. This work demonstrates a tight coupling between the behavior of the subjects and the state of the sensory cortex perturbed by ICMS.

Third, I present work aimed at providing experimentalists with a set of tools for exploring changes in the dynamics of recorded neural ensembles. Despite the growing use of multi-electrode recording arrays, most data analyses are still univariate. This fact is a product of the peculiar properties of neural data as well as the sampling problems associated with trying to relate models of neuronal interactions to behavior. Here I present a novel approach combining a spatial partitioning scheme developed for monitoring streaming telecommunications data with Bayesian statistics and elements of information theory to track changes in the dynamics of neural ensembles on time-scales comparable with behavior. I demonstrated the performance of this method upon simulated ensembles with prescribed properties. Its utility was also made clear by applying it to data collected from rodents.

The major contributions of this work may be divided into the areas of neuro-engineering, empirical studies and applied mathematics. It is my hope that the focus upon learning behavior presented here suggests an integration of these distinct aspects of neuroscience toward understanding the intriguing ability of intelligent systems to learn. Many technical and conceptual problems must still be addressed before a coherent theory of the neural basis of learning behavior may be found. Yet the practice of completing this thesis has left me optimistic.